Design System Convergence — the organ registry, its five verdicts, and the fingerprint ledger that catches drift while it happens

What this document is. The master list of organs — recurring UI concepts — across the augment-it federation, each with its implementations, a fingerprint, and a verdict. Maintained by Converge The Federated Design System.

What it is not. A component library. A style guide. A mandate. Members build what they want; this document records what got built more than once and what the platform team decided about it.

Why Care?

augment-it is heading toward one repo per app, owned by teams of one to three. The architecture is already shaped for it — sovereign members, no shared federation runtime, independent deploys.

The moment a member leaves the monorepo, cross-member duplication stops being greppable. Everything not converged before the split is converged never. This registry is the artifact that has to exist before the first member moves out.

It also encodes the governance stance, which is deliberately permissive:

The difference between purposeful variation and drift is not a property of the code. It is whether a human wrote it down. Two members deliberately diverging is healthy, and gets recorded as sanctioned. The same two members diverging by accident is drift. The code is identical in both cases. Only this registry distinguishes them.

Entry schema

organ: ConnectorChip              # canonical PascalCase name
intent: >                         # one sentence — the user problem, not the markup
  A chip representing an available connector, selectable, showing enabled state.
state: candidate                  # see states below
verdict: PROMOTE                  # PROMOTE | CONVERGE | SANCTION | DEMOTE | WATCH
similarity: identical             # identical | near | structural | intent-only
implementations:
  - { member: pack-runner,       file: apps/pack-runner/src/ConnectorChip.svelte,       lines: 0, form: component, hash: "" }
  - { member: response-reviewer, file: apps/response-reviewer/src/ConnectorChip.svelte, lines: 0, form: component, hash: "" }
best_current: pack-runner         # the strongest implementation, and the promotion source
token_purity: clean               # clean | minor | violations — F8/F11 posture
a11y: documented                  # documented | partial | absent
first_seen: 2026-07-30            # when this organ entered the registry
last_measured: 2026-09-13
fingerprint_history:
  - { date: 2026-07-30, state: identical }
  - { date: 2026-09-13, state: identical }   # 6 weeks, no drift
refactor_doc: null                # context-v/refactors/... when a verdict creates work
gh_issue: null

States

The five verdicts

Defined in full in Converge The Federated Design System. In brief: PROMOTE (3+ members, stable, token-pure, a11y contract) · CONVERGE (should be one dialect, not ready to graduate) · SANCTION (the needs genuinely differ) · DEMOTE (shared-ui primitive with <2 consumers) · WATCH (evidence insufficient — itself a finding, not a deferral).

The published surface today

packages/shared-ui — 2 components, against 79 .svelte files across 19 members. That ratio is the whole problem in one line.

Standing candidates — measured 2026-09-13

These four were recorded in the root DESIGN.md on 2026-07-30 as byte-identical promotion candidates. Re-measured six weeks later, by hash:

Re-measured by git show at four points in history — birth, 2026-07-30, and HEAD:

The finding that justifies the ledger — and it is not the one we expected.

The first reading of this table was "two of the four drifted while sitting in a queue meant to catch drift." That was wrong, and the truth is worse. ColumnMapper and RecordCard were never byte-identical at any commit in the repository's history. Both were copy-pasted at birth from a sibling — the copy happened in someone's clipboard, not in version control — and each was already carrying its own types, prefix and field list in its very first commit. Neither file has changed since 2026-07-30 at all.

So nothing drifted. The list was wrong on the day it was written, because "byte-identical" was asserted rather than measured, and for six weeks every reader inherited the error — including the first draft of this document.

A hand-written measurement is a claim wearing the costume of evidence. The fix is not "re-measure more often"; it is never hand-write the measurement. pnpm organ:drift generates this table, which is why it can be trusted and why the prose above it can not.

Organ registry

Populated by the first sweep (2026-09-13). Scanners cover four member clusters; entries merge here after cross-cluster deduplication.

Sweep 1 complete. Four read-only scanners across 19 members + the platform layer. 24 organs identified; the 17 with cross-member evidence are registered below. Counts are measured, not estimated.

PROMOTE — evidence is sufficient

ConnectorChip pre-promotion fixes (≈2 hours, not a redesign): color: white ×2 → var(--color-on-accent); role="menu" declares no role="menuitem" children (invalid — copy chat/ChatSurface.svelte, the one correct ARIA menu in the product); add aria-haspopup/aria-expanded (it is a menu button); add the first :focus-visible consumer of --focus-ring. Also box-shadow literal → var(--fx-popover-shadow), which exists and is used by nobody.

CONVERGE — should be one dialect

SANCTION — differ on purpose

WATCH — evidence insufficient

EmptyState (best copy: chat, search-results — the only ones naming the action that fills them) · SpinnerIndicator (3 copies, 3 keyframe names, 3 durations — two on screen rotate out of sync) · CardChrome · CollapsibleDisclosure (record-collector is the only one setting aria-expanded) · FireAndForgetQueue · CrossRemoteHandoff ⚠️ (enhanced-records-list omits the localStorage step the other two treat as mandatory — a latent bug, not a design variant).

BLOCKED on the platform, not on agreement

Do not promote these yet. A promoted Button must pick one radius and one padding. Picking them from literals rather than a named scale makes one member's accident into federal law. The token families come first.

What the sweep found that nobody asked about

Three findings surfaced by all four scanners independently, none of which was in any scanner's brief:

1. 170 declarations across 10 members reference tokens that do not exist. --space-*, --radius-* and --z-* have zero declarations in packages/theme. Members invented the names they expected and let the CSS fallback carry the value. → The Federal Layer Never Shipped Space Radius Or Z

2. sort-filter-lens leaks 61 unnamespaced classes, and .error / .row / .muted collide with 16 / 19 / 14 other surfaces today. → The Sort Filter Lens Containment Breach

3. Zero of 19 members have a DESIGN.md. All 19 doc: paths in the federal registry are dangling references. design-index.json and every .design-context.json — the documented agent entry points — do not exist either. The portal reads none of them: docs-portal has no glob and no markdown renderer, and its member list is a hand-maintained array with one entry.

The pattern under all three. The federal layer documented a vocabulary, a per-member doc layer, and an agent context contract. Members complied with what existed and improvised the rest. Nothing ever compared the contract to the runtime — design-drift.mjs compares CSS to CSS, so a token members consume and the theme never declares is invisible to it.

This is the same disease as Structural Invariants Live In Prose So Sweeps Stop Halfway, in a different layer, and it has the same fix.

The push channel already existed — informally, with nowhere to go

The strongest evidence for the governance model is that teams were already flagging their own duplication in code comments:

• search-results/ResultsAccept.svelte: "copy-adapted from search-and-add (spec D7: per-remote copies, no shared runtime; knowingly more fuel for the component library, gh #22)"

• search-and-add/ProviderPalette.svelte: "Borrowed shape: apps/pack-runner's ConnectorPalette/ConnectorChip"

• affiliation-rating-resolver/app.css: "trimmed subset of person-db-resolver's pdr-* sheet"

Nobody was careless. Three teams named the file they copied and the debt they were taking on, in the commit, at the time. There was simply no registry for that declaration to land in, so it stayed in a comment nobody aggregates.

That is what this document is for. The push channel does not need to be built — it needs somewhere to write to.

Known organ seeds

From No Component Library UI Improvised Not Component Based (2026-07-24), unverified at the time and now under sweep:

• WS status pill + client badge header — every remote, copied N times

• Additive URL list with ➕ form — org-workbench (AdditiveList), affiliation-rating-resolver (four lists), person-enrichment (LinkList)

• Candidate picker with score + match_reason — record-db-resolver, person-db-resolver, org-workbench (AddPersonInline)

• Connector/provider chip palette — pack-runner, response-reviewer, search-and-add

• Debounced autocomplete — person-enrichment (org picker), person-db-resolver, org-workbench (OrgSearch)

• Result/source row with one-click action — response-reviewer, corpora-curator (SourceList), search-and-add (ResultRow)

Plus the federation-wide measurement that started this: 158 button rule-sets and 34 badge treatments, none of them components.

Related

• Converge The Federated Design System — the loop that maintains this

• No Component Library UI Improvised Not Component Based — the seed

• Sweep Local Federated Design System for Fidelity — the token lane

• DESIGN.md §The promotion path · §The federation contract · §Local design systems